Losing access to Claude Code - just imagine it. That's really the main reason people push for local open-source models. If you build everything around a closed cloud API, you're basicly staking your whole career on someone else's rules. Anthropic can change prices or shut down your account anytime, and suddenly, your productivity takes a nosedive. There's only one way out: leran to run your own setup. Right now, getting Claude Code to work well alongside Ollama or Qwen is tough. Context management and system instructions are tricky. But honestly, it's our only real safety net. If you can fire up quantized Qwen 2.5 coder model on your own machine, you control your workflow. Sure, it might not be as fast as Sonnet, but there's no kill-switch, no subscriptions, and no strings attached. Think of AI tools like servers: if you can't self-host a backup, you don't really own your workflow. I worked with a friend on a local AI-Agent that can code, that can talk with you about anything and can control your whole computer, even spotify it can easily search on spotify or you can say "open zapzap" or "open whatsapp" it looks for apps on your computer if nothing found with the name it opens the browser and google if you say "open youtube" it opens the browser with https://www.youtube.com. of course you can say it with your voice what you want, so you haven't write it to it per keyboard if your busy. if you wanna try just write a response here and I talk to my friend if I can make it open source. Sorry for self-promoten here but it's something about the topic. if you have a problem with the self-promoten just write a reply and I'll remove this comment.
Hopefully by then local models will be more useful or more accessible on decent hardware.
For now, all my Codex projects include instructions to comment all code for human readers, develop unit tests, and document everything in human readable formats with a human audience in mind.
Caveat: I was a pretty decent programmer before AI.
I've had a Claude max plan that I've been using daily for about 6 months now. This month I decided to turn off auto-renew and moved over to Codex to try it out. There's virtually no switching costs. Of course there are some slight differences I'm noticing between the two but nothing that would keep me loyal to one over the other.
Also, Claude Code is the agentic harness with the underlying models being the anthropic ones. If it's the harness you're worried about, try out Codex or any of the open source ones like OpenCode, Pi, Goose.
This will never go away. Instead it will become more local. In the future, it's more likely that you have a model running on your computer. These companies have no moat. You don't need to worry about it going away or them charging beyond what you can pay.
You can ask Claude to run a pre-mortem on this scenario happening and get it to recommend pre-emptive solutions.
I irregularly use a framework similar to the one by Klein, Sonkin, Paul Johnson https://leonlins.com/writing/premortem to get one AI agent to consider the problems and potential solutions to vendor lock in.
If this is for Claude specifically, feels like most common alternatives (Codex) are viable.
If this is for AI agents generally, I've been backing up local model workflows as well
Yeah, I know how you feel. When my tokens run out doing work, I bounce to YouTube until I'm recharged. Claude, like prostitution, drugs, and gambling, will never go away. That's not how markets work.
If Ed Zitron is to be believed, Anthropic, and all the other LLM based tools companies are bleeding cash at one of the highest burn rates in history, subsidizing your tools, and likely giving them to you at around 5-10% of their costs. Even if you're prepared to pay 10x or more for the same thing, most aren't, and they don't have a sustainable future.
Your best bet is to develop a local capability on your own hardware while you've still got Claude to compare it to.
Second best is to have an LLM model and harness you can run in the cloud independent of Anthropic or any other AI company.
I disagree. I abandoned Claude during Opus 4.8 for GLM 5, it was a flat upgrade for my work. I can't imagine going back to Claude and dealing with refusals or silent model degradation.
No - Here is what I mean. When I use Claude Code locally with Ollama/Quen, it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file. So far I can't convince it to actually WRITE to a file. With Claude Code, it just works. Recursing around the file system, it just does it. I can't get that working locally.
The feature you're describing is called "tool calling" in harness parlance. It's not Anthropic-exclusive, it's probably broken on your local models because you misconfigured Ollama (or it misconfigured itself; Ollama isn't a great inference server).
Losing access to Claude Code - just imagine it. That's really the main reason people push for local open-source models. If you build everything around a closed cloud API, you're basicly staking your whole career on someone else's rules. Anthropic can change prices or shut down your account anytime, and suddenly, your productivity takes a nosedive. There's only one way out: leran to run your own setup. Right now, getting Claude Code to work well alongside Ollama or Qwen is tough. Context management and system instructions are tricky. But honestly, it's our only real safety net. If you can fire up quantized Qwen 2.5 coder model on your own machine, you control your workflow. Sure, it might not be as fast as Sonnet, but there's no kill-switch, no subscriptions, and no strings attached. Think of AI tools like servers: if you can't self-host a backup, you don't really own your workflow. I worked with a friend on a local AI-Agent that can code, that can talk with you about anything and can control your whole computer, even spotify it can easily search on spotify or you can say "open zapzap" or "open whatsapp" it looks for apps on your computer if nothing found with the name it opens the browser and google if you say "open youtube" it opens the browser with https://www.youtube.com. of course you can say it with your voice what you want, so you haven't write it to it per keyboard if your busy. if you wanna try just write a response here and I talk to my friend if I can make it open source. Sorry for self-promoten here but it's something about the topic. if you have a problem with the self-promoten just write a reply and I'll remove this comment.
I've moved to Codex lately, Astra does better with less bugs.
When GPT does poorly, I use Cursor. The autocomplete itself is extremely good.
And then there's Kimi code and similar which can't keep up with the existing demand, much less the demand if all other tools go down.
Then Opencode, which would probably be a year or so behind SOTA but it's good enough.
Hopefully by then local models will be more useful or more accessible on decent hardware.
For now, all my Codex projects include instructions to comment all code for human readers, develop unit tests, and document everything in human readable formats with a human audience in mind.
Caveat: I was a pretty decent programmer before AI.
I've had a Claude max plan that I've been using daily for about 6 months now. This month I decided to turn off auto-renew and moved over to Codex to try it out. There's virtually no switching costs. Of course there are some slight differences I'm noticing between the two but nothing that would keep me loyal to one over the other.
Also, Claude Code is the agentic harness with the underlying models being the anthropic ones. If it's the harness you're worried about, try out Codex or any of the open source ones like OpenCode, Pi, Goose.
This will never go away. Instead it will become more local. In the future, it's more likely that you have a model running on your computer. These companies have no moat. You don't need to worry about it going away or them charging beyond what you can pay.
You can ask Claude to run a pre-mortem on this scenario happening and get it to recommend pre-emptive solutions.
I irregularly use a framework similar to the one by Klein, Sonkin, Paul Johnson https://leonlins.com/writing/premortem to get one AI agent to consider the problems and potential solutions to vendor lock in.
If this is for Claude specifically, feels like most common alternatives (Codex) are viable.
If this is for AI agents generally, I've been backing up local model workflows as well
bliss, that is what happens when it goes away.
“I haven't had great success making Claude Code work with local Ollama/Quen.”
What’s been the biggest blocker for you?
Yeah, I know how you feel. When my tokens run out doing work, I bounce to YouTube until I'm recharged. Claude, like prostitution, drugs, and gambling, will never go away. That's not how markets work.
All bleeding stops.
If Ed Zitron is to be believed, Anthropic, and all the other LLM based tools companies are bleeding cash at one of the highest burn rates in history, subsidizing your tools, and likely giving them to you at around 5-10% of their costs. Even if you're prepared to pay 10x or more for the same thing, most aren't, and they don't have a sustainable future.
Your best bet is to develop a local capability on your own hardware while you've still got Claude to compare it to.
Second best is to have an LLM model and harness you can run in the cloud independent of Anthropic or any other AI company.
use codex
> Someday it will probably go away.
It won't. There are dumps of the early Claude Code codebase/repo and it can be rebuilt from scratch by the community.
You might lose access to Anthropic's models one day, but there are hundreds of alternative model choices.
The model and the harness go hand in hand, in my opinion; other models aren't interchangeable here for quality results.
This is what I'm referring to - the seamless integration - "it just works".
I disagree. I abandoned Claude during Opus 4.8 for GLM 5, it was a flat upgrade for my work. I can't imagine going back to Claude and dealing with refusals or silent model degradation.
I would live as I did in :gasp: 2019. I would simply write code myself and spend less time talking to chatbots. I was more in demand then too.
You mean the model not the harness right
No - Here is what I mean. When I use Claude Code locally with Ollama/Quen, it can do the work just fine - like creating a script - but it just outputs it, rather than placing it into a file. So far I can't convince it to actually WRITE to a file. With Claude Code, it just works. Recursing around the file system, it just does it. I can't get that working locally.
The feature you're describing is called "tool calling" in harness parlance. It's not Anthropic-exclusive, it's probably broken on your local models because you misconfigured Ollama (or it misconfigured itself; Ollama isn't a great inference server).
[dead]
[flagged]
[flagged]
[dead]
[dead]
[dead]
[dead]
[dead]